g1_dex3_pickgum_dataset (TsFile)

Apache TsFile version of unitreerobotics/G1_Dex3_PickGum_Dataset.

Overview

A LeRobot demonstration dataset recorded on a Unitree G1 humanoid equipped with Dex3 dexterous hands. Each record is one control frame: the commanded action, the measured robot state, and the frame/task bookkeeping for a single manipulation episode. The dataset covers one pick-and-place task performed by the dual-arm/hand system.

• Task: "Put the gum into the plate."
• Scale: 199 episodes, 113,592 frames at 30 fps, 1 task.
• Robot: Unitree_G1_Dex3 (28-DoF action and observation state: both arms plus both Dex3 hands).
• Modalities converted: the 28-D scalar observation.state and 28-D scalar action, plus frame_index/sample_index. The two camera video streams (observation.images.cam_left_high, observation.images.cam_right_high) are not included; see below.

Schema (TsFile structure)

• Time (INT64, milliseconds) — round(timestamp * 1000), restarting at 0 for each episode.
• episode_index (TAG) — episode/device dimension. Query one episode with WHERE episode_index=0.
• task_index (TAG) — task id (single task, always 0).
• frame_index, sample_index (INT64) — source frame number and global source row index.
• action_0 .. action_27 (FLOAT) — commanded arm and hand joint positions.
• observation_state_0 .. observation_state_27 (FLOAT) — measured arm and hand joint positions.

Vector columns are flattened element-wise (observation.state → observation_state_0..27, action → action_0..27). The source timestamp column is not retained because it equals Time ÷ 1000 seconds.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("data/g1_dex3_pickgum_dataset.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())

Source & license

• Original dataset: https://huggingface.co/datasets/unitreerobotics/G1_Dex3_PickGum_Dataset
• Author / publisher: unitreerobotics
• License: apache-2.0
• Videos: the source camera videos are not uploaded to this repository. See the original dataset's videos/ directory.